[Bug target/38026] New: Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[Bug target/38026] New: Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu dot org]

To repeat, run gfortran.dg/char_result_5.f95 from th 4.3-branch with -fno-gcse.
In the gcc objdir:
make check-fortran 'RUNTESTFLAGS=--target_board=cris-sim/-fno-gcse
dg.exp=char_result_5.f90'

Observe:
Using /tmp/hpautotest-gcc43/gcc/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /tmp/hpautotest-gcc43/gcc/gcc/testsuite/gfortran.dg/dg.exp ...
FAIL: gfortran.dg/char_result_5.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: gfortran.dg/char_result_5.f90  -O3 -fomit-frame-pointer
-funroll-all-loops -finline-functions  execution test

with gfortran.log saying for both FAILs:
PASS: gfortran.dg/char_result_5.f90  -O3 -fomit-frame-pointer -funroll-loops 
(test for excess errors)
Operating system error: Illegal seek^M
Memory allocation failed^M
FAIL: gfortran.dg/char_result_5.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test

I'm guessing a target-bug, exposed by -fno-gcse.


-- 
           Summary: Miscompilation for CRIS of gfortran.dg/char_result_5.f95
                    with -fno-gcse
           Product: gcc
           Version: 4.3.3
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: hp at gcc dot gnu dot org
  GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: cris-axis-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

[Bug target/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu dot org]

-- 

hp at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |hp at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-11-05 18:27:32
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

[Bug rtl-optimization/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu dot org]

------- Comment #1 from hp at gcc dot gnu dot org  2008-11-06 02:06 -------
No, it looks like (sigh) a delayed-branch bug; a delay-slot is filled with a
value from a forward branch clobbering the return value from a call. 


-- 

hp at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |rtl-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

[Bug rtl-optimization/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu dot org]

------- Comment #2 from hp at gcc dot gnu dot org  2008-11-06 03:05 -------
Sure, it's fill_eager_delay_slots that moved the insn, but it's only acting on
a stray REG_DEAD note left over from cprop_hardreg.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

[Bug rtl-optimization/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |WAITING

--- Comment #3 from Steven Bosscher <steven at gcc dot gnu.org> 2013-04-18 21:53:25 UTC ---
(In reply to comment #2)
> Sure, it's fill_eager_delay_slots that moved the insn, but it's only acting on
> a stray REG_DEAD note left over from cprop_hardreg.

This is probably fixed by http://gcc.gnu.org/r164458

[Bug rtl-optimization/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|                            |WORKSFORME
   Target Milestone|---                         |4.3.6

--- Comment #4 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2013-04-19 10:18:32 UTC ---
(In reply to comment #3)
> (In reply to comment #2)
> > Sure, it's fill_eager_delay_slots that moved the insn, but it's only acting on
> > a stray REG_DEAD note left over from cprop_hardreg.
> 
> This is probably fixed by http://gcc.gnu.org/r164458

That might be so, but I can't verify that. :(

Unfortunately I forgot to note the exact svn revision at which the test failed,
and reverting the 174133 revision (backport to 4.3 documented in PR42775) from
revision 175148 and following the to-repeat description does not expose the
bug.

Hm, I see I mentioned 4.3.3 in this quite old report but as I can't repeat it
with a revision near ToT, I believe it's time to retire this PR and close it as
WORKSFORME rather than WONTFIX.

[Bug rtl-optimization/38026] Miscompilation for CRIS of gfortran.dg/char_result_5.f95 with -fno-gcse

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38026

--- Comment #5 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2013-04-20 01:54:59 UTC ---
(In reply to comment #4)
> I can't repeat it
> with a revision near ToT,

I tested using r175148 on the 4.3-branch, in case it was unclear.